I’ve been batting about an essay for two months now on the question, “Should you work for Neuralink?” However, I’ve ultimately failed to answer this question to my satisfaction. So, I thought I’d explain how far I got, where I’m stuck, and then what I would need to know in order to actually answer this question.
So, first of all, context: Neuralink is a brain-computer interface company run by and founded by Elon Musk. It was launched in 2016, intermittently gets a lot of hype, and seems to bleed employees, including 6 out of 8 of the original scientific cofounders.
Thanks for reading my newsletter! Subscribe for free to receive new posts, which are usually much more finished than this.
I never really used to care about Neuralink, to be honest. First, I’m naturally skeptical of all companies founded on highly speculative technologies, and I think most of them should be nonprofit research projects. Second, I’m skeptical of most Elon Musk projects, as it’s exhausting separating the hype from the reality. Third, and most importantly, I never really got why anyone would care about brain-computer interfaces, as I didn’t get any practical purposes of linking together brains and machines outside of very niche applications.
What prompted this essay, however, is that I recently changed my mind on this last point. A couple months ago, I played an excellent cyberpunk videogame, Citizen Sleeper. Like any good cyberpunk media, Citizen Sleeper grapples a lot with societal and personal benefits and consequences of near-future tech. One of the pieces of near-future tech that comes up consistently is the brain-computer interface (BCI).
Playing Citizen Sleeper made me realize some of the promises of BCI, as well as think more thoroughly about the downsides. In the game, you meet many characters who have either chosen or been forced to use BCIs. If you’re ok with some spoilers, in Citizen Sleeper you meet:
[Start spoilers]
a) a paralyzed space captain who uses a brain computer interface to control a suit of robot armor that pilots his body
b) a short-order cook who still bears scars from where his BCI-piloted robot was welded to his body by a faceless corporation
c) a gangster who uses a BCI to control a superstrong robot arm, only to find out that the robot arm can also be remotely controlled by the manufacturer
Even you yourself, the titular “Sleeper”, are a product of a BCI. You rented your own consciousness to a corporation, who uploaded it through a BCI to an artificial body that they used as slave labor while the original body remained in stasis.
[End spoilers]
These are all semi-realistic uses of BCIs. I get why these might be appealing in industry or in personal life.
Meanwhile, around the same time that I was completing Citizen Sleeper, Elon Musk’s Neuralink, a BCI company, had a technical presentation. Like all Elon Musk’s stuff, this came with a fair amount of hype. As of writing, his presentation has over 600,000 views, which is pretty amazing for a 2 hour, 45 minute dry technical presentation.
At the end of Elon Musk’s introduction, he talks about how this presentation is specifically a recruiting presentation. He says if you’ve ever had any part in designing any kind of electronic hardware, you should apply to work at Neuralink.
So, I wanted to determine for myself if that was the case. Should someone technically skilled apply to work for Neuralink?
Well, to my mind, before you can answer that question, there would actually be 5 questions you’d have to answer.
1. Should you be working in BCI at all?
2. Are Neuralink’s goals valuable?
3. Does Neuralink’s strategy make sense?
4. Is Neuralink’s track record impressive?
5. Are there ethical issues in working for Neuralink?
I tried to take these questions one-by-one. Here’s how far I got.
Should you be working in BCI at all?
Depending on how seriously you take a sci-fi future, this question might be easy to answer for you. The quasi-dystopian future of Citizen Sleeper would not be possible without serious advancements in BCIs, nor would Elon Musk’s wild ideas of using BCIs to make sure humans reach parity with eventual superintelligent AIs.
However, to my mind, you could make the same sci-fi future argument about flying cars, but I think you would have had a really miserable career if you dedicated yourself to flying cars for the past 50 years. So, I think a better question would be, “If you dedicate the next 50 years of your career to BCIs starting now, will you have a satisfying career?”
This is a complex question. Here’s one way of looking at it. If you had started working in BCIs 25 years ago, here would have been the possible highlights on your resume:
1) Reproducing the images seen by cats on a computer screen by reading their brain (1999)
2) Helping blind people see again (2002)
3) Teaching monkeys how to control robot arms (2013) using only their brain
4) Allowing tetraplegics to control robot prosthetic limbs (2014)
5) Giving monkeys sensory feedback after controlling robot arms directly to their sensory cortex (2017)
6) Allowing ALS patients to operate computers (2020)
7) Synthesizing speech for patients with neurological disorders (2022)
These would all be cool highlights to have! It’s not necessarily the most obvious order (who would have guessed that synthesizing speech would come 20 years after sight), but it would still be neat.
One obvious drawback that you might notice to working on these is that none of these are commercialized. The average blind person can’t get any sort of implant to see again 20 years later (although there are a lot of recent studies into this). Same thing with walking or controlling robot limbs for the average paraplegic or tetraplegic. So far, blind people still rely on seeing eye dogs, and paraplegics still rely on wheelchairs and long sticks. So, these highlights would not have produced any significant real-world impact.
Now, it seems pretty obvious as to why these haven’t produced any real world impact. That’s because all these interventions are expensive, one-off solutions that require top-notch surgeons, electrical engineers, and programmers. It’s very tough to scale stuff like that.
But, as you might imagine, people are always claiming we are very close to getting to a place where we can get easy-to-implant, mass-manufactured BCIs that would be widely available. This includes stuff like Neuralink’s wireless transmission and the Stentrode, which is a BCI that’s located in the jugular vein rather than in the brain itself (i.e. it doesn’t require open brain surgery).
I just lack the expertise to say whether we can legitimately expect these innovations to lead to safer, cheaper, easier-to-implant BCIs, even if all the plans of the founders go exactly as promised. So that’s the first place I got stuck: I don’t really understand the challenges in mass-producing any of this.
Are Neuralink’s goals valuable?
This one’s easier. I think, by and large, Neuralink’s stated goals aren’t valuable.
They say on their website, “We’re aiming to design a fully implantable, cosmetically invisible brain-computer interface to let you control a computer or mobile device anywhere you go”.
I don’t think most people want this or need this. Unlike Elon, I’m fine using my fingers to scroll Twitter as I don’t need the timeline downloaded directly to my brain. Also, unlike Elon, I don’t think connecting human brains to supercomputers is a good or realistic goal, and I don’t really see a clear path from what they’re doing now to robot brains.
However, a mass-market BCI could be great for blind or paralyzed people. So, maybe we can expect that this supercomputer nonsense ends up having an ulterior benefit. I do worry that, if Neuralink does get on the market, they’d bank so much on the supercomputer crowd that the product would flop and Neuralink wouldn’t even end up being available to blind or paralyzed people.
Does Neuralink’s strategy make sense?
Neuralink has a multi-pronged scientific strategy for achieving their scientific goals that’s difficult for me to evaluate scientifically, but I can say confidently that it requires a frighteningly lot of things to go right.
They plan to have micron scale threads that attach to an implant that transmits information about neural impulses wirelessly. These micron scale threads and the implant will be resistant to corrosion from brain fluid and last for years. The threads and implant themselves will be installed via robot surgery.
All of these pieces of tech are complicated and would require internal development on the part of Neuralink to work correctly. None of them are off-the-shelf. Getting all these things working correctly and working together is hard!
If you listen to Neuralink’s presentation, they act like they’ve already got every part of this working perfectly in monkeys. However, there’s really no way of verifying exactly what it is that they’re doing, as they haven’t published anything, as far as I know.
I am, again, not well versed enough in materials engineering, electrical engineering, or robotics to really evaluate exactly how difficult all of this is. But this does seem like a lot of things that will need to be developed. I generally don’t think it’s a good idea for a new team to make a new product that requires a bunch of technical advances, as usually just getting existing things to work together is hard enough.
Is Neuralink’s track record impressive?
As I mentioned, I can’t really know exactly what’s going on with Neuralink’s monkey demos. What seems clear is that they’ve developed some sort of wireless interface that allows monkeys to play pong or to select letters from a keyboard.
The only part of this that’s more impressive than the demos from 10 years ago is the wireless part. Again, scientists taught monkeys how to control robot arms in 2013. That’s at least on the same level as selecting letters from a keyboard.
Wireless is definitely some kind of an advance, but it’s hard for me to say exactly how impressive it is because I don’t know enough about wireless. My only real point of comparison is the Oculus VR headset, which is still the only wireless headset after 4-ish years even though that’s an obvious selling point. My guess from that is that wireless is hard and expensive to implement in hardware, so it’s impressive that Neuralink managed it. Their other accomplishments are less impressive.
Are there ethical issues in working for Neuralink?
There are always weird issues working for Elon Musk’s companies. He’s a weird guy and there’s always negative press.
Most recently, there was a report in 2022 that, from 2017 to 2020, 15 out of 23 monkeys were euthanized as part of Neuralink’s experiments. It’s also possible some of those monkeys underwent severe psychological distress.
My feelings on this are mixed. If I was convinced that Neuralink was clearly heading in the direction of, say, giving blind people sight, I might think this is an ok price to pay. But, as you can probably tell, I’m not convinced, so this does kind of seem like torturing monkeys without a good reason as to why. That seems bad.
Conclusion
Overall, I would lean against working for Neuralink, but I’m really not confident on this. Before answering this definitively, I’d want to know:
1. The reason why BCIs aren’t commercially available 20 years later, including a better idea of the market size, the marginal cost of developing and installing them, and the challenges of manufacturing.
2. How likely it is that one of the other approaches to BCIs (i.e. the Stentrode) will lead to better results.
3. The engineering challenges behind Neuralink’s exact approach.
And that’s where I’ll leave this, unless I come back to this in several months because I’m still bothered by the unfinished nature of this post.